"The Cognitive Revolution"

The episode delivers several genuinely useful datapoints on how AI compresses engineering design cycles, including the workflow mechanics and the F1 compute-cap governance angle, but the signal is diluted by the host's lengthy, meandering preambles and repeated software/protein-folding analogies that consume significant runtime without adding new substance.

“from 50 designs evaluated per day to 1,500 every single day in production”

“these models are always evolving over time in the sense that every time you feed them your data out, they are being retrained and improved so that they can cover a, uh, broader and broader space and become more and more accurate”

The core thesis - AI compresses simulation time by orders of magnitude, enabling exponentially more design iterations - is a well-worn pattern in AI podcasting; the guest applies it competently to engineering but adds little truly counterintuitive framing. The most genuinely fresh angle is the Formula One CPU-hour handicapping system as a proto-model for AI compute governance, but even that is not developed into a strong original argument.

“depending on your ranking from the previous year, you don't get the same numbers always for the next season. The idea is that you want to try and make it more equal across teams”

“engineering is such a specific, accurate field that there is not one foundational model that can be used to solve the aerodynamics on every car”

Thomas von Tschammer is a genuine practitioner - co-founder actively deploying these systems at named enterprise clients like JLR and Formula One teams - which distinguishes him from thought-leader guests; however, he is a startup founder/BD figure rather than a chief engineer or simulation scientist, and several answers stay at a strategic level rather than revealing deep technical specifics.

“They published their work with us uh, at the latest Nvidia GTC conference that was back in March this year”

“We started in 2019 by building the first AI model architecture based originally on computer vision that could directly ingest geometries and learn from physics”

The episode is anchored by a handful of concrete, named metrics - JLR's 50-to-1,500 designs per day, the 48-60 versus 18-24 month OEM development cycle gap, 80% cycle reduction for battery cold-plate suppliers - and specific named entities (JLR, F1 teams, Nvidia GTC); however, many other claims (2-3-5% aero gains, year-1 and year-2 speedup projections) are stated without sourcing, and supplier examples remain anonymous.

“they went from 50 designs evaluated per day to 1,500 every single day in production”

“Western Europe or US OEM, it takes them between 48 to 60 months for a new car developed... in China it's 18 to 24 months”

The host is well-prepared and occasionally lands a genuine push - asking 'Are you sure?' on the question of whether AI will replace engineers and demanding the guest defend the ceiling - but most questions are long, answer-telegraphing monologues that let the guest coast, and there is minimal follow-up pressure when the guest retreats to safe generalities about 'trade-offs' and 'domain experts in the loop.'

“Are you sure? What does a Bake off look like there?”

“If that's crazy, why do you think that's crazy? What is the part that we're so far off on?”